Land Use Regression Models for Alkylbenzenes in a Middle Eastern

Article

Land use regression models for alkylbenzenes in a Middle Eastern megacity: Tehran Study of Exposure Prediction for Environmental Health Research (Tehran SEPEHR) Heresh Amini, Christian Schindler, Vahid Hosseini, Masud Yunesian, and Nino Künzli Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.7b02238 • Publication Date (Web): 28 Jun 2017 Downloaded from http://pubs.acs.org on June 30, 2017

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Environmental Science & Technology is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 28

Environmental Science & Technology

1

Land use regression models for alkylbenzenes in a Middle Eastern megacity:

2

Tehran Study of Exposure Prediction for Environmental Health Research (Tehran

3

SEPEHR)

4

Heresh Aminia,b,*, Christian Schindlera,b, Vahid Hosseinic, Masud Yunesiand and Nino

5

Künzlia,b,*

6

a

7

Institute, Basel, Switzerland

8

b

University of Basel, Basel, Switzerland

9

c

Mechanical Engineering Department, Sharif University of Technology, Tehran, Iran

10

d

Center for Air Pollution Research (CAPR), Institute for Environmental Research (IER),

11

Tehran University of Medical Sciences, Tehran, Iran

Department of Epidemiology and Public Health, Swiss Tropical and Public Health

12 13

*Corresponding authors:

14

Heresh Amini

15

Tel: +41 61 284 8932

16

E-Mail: [email protected]

17

AND

18

Prof. Dr. Nino Künzli

19

Tel: +41 61 284 8399

20

Fax: +41 61 284 8105

21

E-Mail: [email protected]

22

Submitted to Environmental Science & Technology

23

Text word count: 4484

1 ACS Paragon Plus Environment


Page 2 of 28

24

ABSTRACT

25

Land use regression (LUR) has not been applied thus far to ambient alkylbenzenes in

26

highly polluted megacities. We advanced LUR models for benzene, toluene,

27

ethylbenzene, p-xylene, m-xylene, o-xylene (BTEX), and total BTEX using measurement

28

based estimates of annual means at 179 sites in Tehran megacity, Iran. Overall, 520

29

predictors were evaluated, such as The Weather Research and Forecasting Model

30

meteorology predictions, emission inventory, and several new others. The final models

31

with R2 values ranging from 0.64 for p-xylene to 0.70 for benzene were mainly driven by

32

traffic-related variables but distance to sewage treatment plants was present in all

33

models indicating a major local source of alkylbenzenes not used in any previous study.

34

We further found that large buffers are needed to explain annual mean

35

concentrations of alkylbenzenes in complex situations of a megacity. About 83% of

36

Tehran’s surface had benzene concentrations above air quality standard of 5 µg/m3 set

37

by European Union and Iranian Government. Toluene was the predominant

38

alkylbenzene and the most polluted area was the city center. Our analyses on

39

differences between wealthier and poorer areas also showed somewhat higher

40

concentrations for the latter. This is the largest LUR study to predict all BTEX species in

41

a megacity.


Page 3 of 28


42

INTRODUCTION

43

Numerous studies have shown that air pollution causes acute and chronic morbidities

44

and premature death1 resulting in considerable public health consequences.2, 3

45

Land use regression (LUR) models have been extensively used for long-term

46

exposure assessment to ambient air pollutants, which is essential for epidemiologic

47

research on chronic health effects of air pollution.4, 5 In LUR, many spatially varying

48

variables are used to predict measured concentrations of air pollutants using

49

regression models. Finally, these explanatory variables can be used to estimate

50

pollution anywhere in the study area.6

51

Air pollution consists of a wide range of particles and gases.7 Among the gases, there

52

is a group of volatile organic compounds (VOCs) classified into alkanes,

53

alkylbenzenes, chlorinated hydrocarbons, terpenes, and other miscellaneous VOCs.8

54

Alkylbenzenes include toxic air pollutants, such as benzene, toluene, ethylbenzene,

55

p-xylene, m-xylene, and o-xylene (BTEX).

56

To date, several LUR studies, mainly from North America, have been published

57

where they measured and estimated exposure to VOCs.9-11 However, these studies

58

were mainly conducted in areas where concentrations have been very low or relevant

59

only for small populations.9-12 LUR has not been applied so far for alkylbenzenes in

60

highly polluted megacities. In a LUR study from New York City,12 covering an area

61

with a population of about 8 million people, benzene, total BTEX, and formaldehyde

62

were modeled based on three months of measurements, at very low concentration

63

levels (e.g., with mean benzene = 0.8 µg/m3). While several studies modeled (m+p)-

64

xylene, none looked at m and p-xylene separately. Furthermore, none of these LUR 3 ACS Paragon Plus Environment


Page 4 of 28

65

studies incorporated meteorological variables, such as wind speed, relative humidity,

66

or temperature. Overall, studies from outside North America and Europe with a wide

67

range of concentrations and different populations are critically needed.13

68

The Tehran Study of Exposure Prediction for Environmental Health Research

69

(Tehran SEPEHR) is a collaborative Swiss-Iranian effort aimed to measure and

70

estimate long-term spatial variability of air pollution—mainly of nitrogen dioxide, sulfur

71

dioxide, ozone, and BTEX—and to estimate chronic exposure of Tehran megacity

72

residents to these pollutants for use in future studies. In line with our previous

73

publications where concentrations of criteria air pollutants were well above the values

74

recommended by the World Health Organization,14, 15 the recent Tehran SEPEHR

75

results demonstrated that ambient concentrations of toxic VOCs were very high and

76

a public health concern in Tehran.16

77

The aims of this study were a) to develop LUR models for all measured

78

alkylbenzenes in Tehran megacity, using common land use and traffic variables but

79

also evaluating some new predictor variables, b) to analyze their spatial variability,

80

and c) to analyze the difference of air quality in advantaged versus disadvantaged

81

areas.

82 83

EXPERIMENTAL

84

Research field

85

Tehran SEPEHR was conducted in Tehran, a mega-city with a resident population of

86

about 9 million people and a day-time population of over 10 million due to massive

87

commute from outer areas.14 The community covers an area of over 613 km2 with 4 ACS Paragon Plus Environment

Page 5 of 28


88

higher population densities in the mid-west and mid-east to southern areas. The

89

Alborz Mountains surround the northern areas of Tehran and there is desert in the

90

south. As a result, there is a difference of 800 m in altitude from the populated areas

91

in the south to those in the north Tehran (Figure 1). The climate is semi-arid with an

92

annual mean temperature of 18.5 °C. Minimum temperatures go down to −15 °C in

93

January and reach 43 °C in July. The sun often shines in Tehran (annual bright

94

sunshine of 2800 hrs and mean cloud cover of 30%) and there is an annual

95

precipitation of about 220 mm. The maximum of 39 mm and minimum of 1 mm occur,

96

respectively, in January and September.15

97

Measured alkylbenzenes

98

The details of research design, methods, and descriptive results of the Tehran

99

SEPEHR on alkylbenzenes are published elsewhere.16 In brief, we first visited and

100

noted the characteristics of 276 eligible measurement sites in consultation with local

101

collaborators. A subset of 174 sampling sites were selected with respect to sampler

102

deployment permission, budget, and various land uses using a cluster analytic

103

method and prior knowledge about spatial distribution of air pollution in Tehran. Most

104

of the sites were in the residential areas as we were interested at general

105

population’s exposure to air pollution. In addition, 5 reference sites were selected to

106

represent a wide range of traffic, elevation, and emissions. BTEX were measured

107

consecutively every two weeks at the 5 reference sites throughout the entire year

108

from April 2015 to May 2016 and at other 174 sites over three two-week periods in

109

summer, winter, and spring seasons by Passam passive samplers (Figure 1). Finally,

110

using the ratios of these measurements to concurrent levels at reference sites,

111

annual mean levels of the different alkylbenzenes were estimated for all sites.16 They

112

defined the response variables of our LUR models. 5 ACS Paragon Plus Environment


Page 6 of 28

113

Predictors

114

We developed 520 predictor variables from a wide variety of features categorized into

115

ten classes and according to different buffer zones. These were traffic, non-traffic,

116

population, geography, land use, distance, log-distance, products, emission, and

117

meteorology (supplemental information (SI), Table S1).

118

The traffic class included lengths (in meters) or counts (#) of the following traffic

119

parameters derived in various buffer sizes from 25 to 5000 m: All roads, highways,

120

main roads, ancillary roads, alleys, all bus lines, bus rapid transportation (BRT) lines,

121

non-BRT lines, taxi lines, bridges, all bus stops, BRT stops, non-BRT stops, taxi

122

stops, and critical traffic points. The non-traffic class included counts of blocks,

123

primary schools, and high schools. The population class included total population

124

density with buffers of 500 to 3000 meters. The geography class included variables of

125

ground slope and elevation, which are derived from a digital elevation model. The

126

land use class included areas of residential, green space, urban facilities, industrial,

127

official/commercial, transportation, sensitive, agriculture, arid or undeveloped, and

128

other land uses within various buffers from 25 to 5000 meters (in m2 units) . All

129

derived distance and log-distance variables are listed in SI, Table S1. The products

130

included the ratio of buffer-based variables in the traffic, non-traffic, and land use

131

classes to the variables in the distance class. The product variables for LUR have

132

been used in previous studies as they might better represent the emission and

133

dispersion processes.17 We used an emission inventory of VOCs as a potential

134

predictor for alkylbenzenes in LUR. The details of the emission inventory estimations

135

can be found elsewhere.18, 19


Page 7 of 28


136

The non-hydrostatic, mesoscale Advanced Research Weather Research and

137

Forecasting model, version 3.4, was used as the meteorological model.20 This

138

mesoscale model is a state-of-the-art atmospheric simulation system based on the

139

fifth-generation Penn State/ The National Center for Atmospheric Research

140

Mesoscale Model.21 Meteorological data over the Tehran domain including minimum,

141

average, and maximum wind speed, relative humidity, and temperature were

142

calculated using the Weather Research and Forecasting (WRF) model with three

143

nested domains having 27, 9 and 3 km resolutions, respectively. The 27-9-3 km

144

domains were run together efﬁciently using two-way grid nesting in WRF. The lateral

145

boundary conditions for coarse domain were taken from the National Centers for

146

Environmental Prediction Global Forecast System (0.5 × 0.5 degree) at six hours

147

frequency. After calculating the daily means, the annual mean of the

148

abovementioned estimated meteorological variables were calculated and used as

149

predictors in LUR.

150

Model building and validation algorithm

151

Our LUR model building algorithm and validation consisted of 10 steps:

152

(1) Log-transform the response variable if they were not normally distributed.

153

(2) Create a box plot for the response variable and remove outlier observations in

154

both ends from the dataset (i.e, observations that are larger than the 75th

155

percentile or smaller than the 25th percentile by at least 1.5 times the

156

interquartile range (IQR)).

157 158 159

(3) Conduct a univariate regression by regressing log-transformed monitored concentrations against each of 520 predictors. (4) Rank the variables based on adjusted R2 of univariate regression analyses.



160 161 162

Page 8 of 28

(5) Build a "start model" with the variable that gives the highest adjusted explained variance (R2). (6) Add the next 519 variables individually and retain them if: (a) Adjusted R2

163

increases by at least 0.5%, and (b) The direction of the effect is as expected,

164

and (c) The directions of effects of predictors already included in the model do

165

not change, and (d) All variables of the model, including the one added, have

166

p-value < 0.1. Repeat the analyses for all remained predictors in each step

167

until no other variable could be added.

168 169 170 171

(7) Calculate the variance inflation factor (VIF) of each variable for the final model and keep those that are less than 10.14, 22 (8) Calculate Cook’s D statistic and remove influential observations that have values above 1 and run the regression again.23

172

(9) Conduct a leave-one-out cross-validation (LOOCV) and report LOOCV R2

173

(10)

174

Calculate Moran’s Index (I) of prediction residuals and report I and its p-

value

175

Our prior experiences (not shown here) demonstrated that constructing our LUR

176

models based on all observations resulted in very large Cook’s D values (> 1),

177

and consequently led to the removal of about 12 observations. These influential

178

sites were outliers and mainly in areas where no population lived (e.g. inside a

179

park or close to a busy highway), therefore we decided to not consider these

180

measurements as these sites do not represent exposure conditions of any

181

population.


Page 9 of 28


182

Mapping of alkylbenzenes

183

Regression mapping of all estimated alkylbenzenes was done according to our

184

previously published method.14, 15 In brief, we mapped the predictions provided by the

185

final regression equations using the Raster Calculator in the ESRI’s ArcGIS 10.2.1 for

186

Desktop (ESRI, Redlands, CA).

187

Spatial analyses of predicted maps

188

The profile of predicted pollutants over a transect of about 28.4 km from north to

189

south of the city was sampled and plotted at a 5x5 m resolution (5681 cells).

190

In addition, the maps were divided into north to center (advantaged) and south to

191

center (disadvantaged) areas and summary statistics including north vs south min,

192

mean, and maximum concentrations for all pollutants were calculated.

193

RESULTS

194

Alkylbenzenes in Tehran

195

The concentrations of all measured alkylbenzenes were not normally distributed. The

196

annual medians (IQR) for measured pollutants in µg/m3 units were as follow:

197

benzene 7.8 (3.6), toluene 23.2 (10.9), ethylbenzene 5.7 (2.3), p-xylene 5.6 (1.9), m-

198

xylene 10.4 (4.4), o-xylene 6.1 (2.7), and Total BTEX 58.6 (24.8) (Table 1). The

199

spearman spatial correlations of measured alkylbenzenes ranged from 0.81 for

200

benzene and o-xylene to 0.98 for ethylbenzene and m-xylene. Benzene was highly

201

correlated with ethylbenzene (spearman rank correlation = 0.96) (SI, Table S2).

202

Detailed descriptive results can be found elsewhere.16



Page 10 of 28

203

Final LUR models

204

All alkylbenzenes were modeled on the log-scale as they were not normally

205

distributed. Overall, 5 observations for toluene and ethylbenzene, 6 for benzene, p-

206

xylene, and m-xylene, and 7 for o-xylene and total BTEX were removed from the data

207

set for modeling. The variable “all roads in buffer of 5000 m” provided the highest

208

adjusted R2 among all 520 predictors (ranging from 0.35 for p-xylene to 0.40 for

209

benzene) and considered as start model for all pollutants except for o-xylene where

210

official/commercial area within a 3500 m buffer qualified as variable to start the model

211

(adjusted R2 = 0.26). In general, across all alkylbenzenes, 38 predictors were

212

present in any of the models with p-values ranging from < 0.001 up to 0.067. Among

213

them many appeared in the models of several or all alkylbenzenes. Of these, 16 were

214

from land use class, 7 from traffic class, 6 from distance class, 4 from log-distance

215

class, 4 from products class, and 1 from non-traffic class. The distance to sewage

216

treatment plants was the only variable that was present in all models with a negative

217

direction of the effect. Official commercial land use areas in a buffer of 2000 m and

218

the ratio of green space areas in a buffer of 5000 m and distance to green spaces

219

were present in five out of the six models; ancillary roads in a buffer of 50 m, log-

220

distance to taxi lines, sensitive land use areas in a buffer of 2000 m, and taxi lines in

221

a buffer of 50 m were present in 50% of models (Table 2). Over all, the number of

222

variables in the final models ranged from 9 for benzene to 15 for o-xylene. WRF-

223

based variables, emission inventory variables, distance to airport, and distance to

224

gas stations did not contribute to the explanation of long-term variability of

225

alkylbenzenes in Tehran. Our models could explain between 64% (p-xylene) and

226

70% (benzene) of the variance of annual alkylbenzenes in Tehran. The average VIF

227

of variables was less than 2.2 in all models. In 5 out of 6 models, the maximum VIF of 10 ACS Paragon Plus Environment

Page 11 of 28


228

individual variables in the models was lower than 3 but in one model (ethylbenzene) it

229

approached 5. One observation in o-xylene and one in total BTEX had Cook’s D

230

values higher than 1 and were thus removed from the respective models. The

231

LOOCV R2 values ranged from 0.59 for p-xylene and o-xylene to 0.66 for benzene.

232

The Moran’s I values ranged from – 0.02 to 0.01 with p-values ranging from 0.38 to

233

0.47 (Table 2).

234

Regression mapping

235

The final LUR models predicted concentrations of benzene, toluene, ethylbenzene,

236

p-xylene, m-xylene, o-xylene, and Total BTEX at 24,505,474 cells of Tehran area

237

(Figure 2 and SI, Figure S1 and Figure S2). The predicted concentrations ranged

238

from 1.9 to 29.0 for benzene, 5.5 to 64.4 for toluene, 1.3 to 17.0 for ethylbenzene, 1.5

239

to 41.0 for p-xylene, 2.4 to 44.0 for m-xylene, 2.0 to 19.0 for o-xylene, and 18.7 to

240

472.0 µg/m3 for Total BTEX. The spatial spearman correlations of predicted

241

alkylbenzenes ranged from 0.78 for benzene and o-xylene to 0.96 for ethylbenzene

242

and m-xylene (SI, Table S3). The agreement of measured versus predicted

243

alkylbenzenes were relatively good (SI, Figure S3).

244

Spatial analyses of predicted maps

245

Overall, about 90% of measured sites had benzene concentrations above the air

246

quality standard of 5 µg/m3 set by the European Union,16 and about 83% of Tehran

247

area had predicted benzene concentrations above this value with maximum values

248

up to 29 µg/m3. The profile of predicted alkylbenzenes from north to south over a

249

distance of 28.4 kilometers showed that toluene was the dominant pollutant among

250

all alkylbenzenes in Tehran followed by m-xylene, benzene, o-xylene, and p-xylene.



Page 12 of 28

251

The lowest concentrations of this profile were in the north and the highest in the city

252

center towards south (Figure 3).

253

Although the maximum concentrations were higher in the wealthier northern to

254

central areas of the city as compared to the disadvantaged southern to central areas

255

(except for toluene and o-xylene), the minimum and mean concentrations were

256

somewhat lower for all pollutants in the northern advantaged areas (Table S1 and SI,

257

Figure S4).

258

DISCUSSION

259

In this study we developed, for the first time, LUR models for alkylbenzenes in a

260

heavily polluted Middle Eastern megacity. To date, there has been no LUR study on

261

alkylbenzenes considering meteorology in the modeling. We further analyzed the

262

spatial variability of the predicted pollutants first over a north south profile and then

263

over north-to-central advantaged versus south-to-center disadvantaged areas of

264

Tehran megacity.

265

Baldasano and colleagues (1998) tabulated several sources for alkylbenzenes in the

266

ambient air, such as vehicles, gasoline vapor, wastewater treatment, wood

267

combustion, and a range of industrial activities.8 The LUR studies on VOCs have

268

mainly focused on traffic-related variables and industrial activities as predictors and

269

none have considered the other sources mentioned by Baldasano et al.8 There are

270

few considerable wood combustion point sources in Tehran but we have included

271

distance to sewage treatment plants as a predictor variable. Interestingly, this

272

predictor was the only one being selected in all LUR models indicating a likely major

273

source of ambient air pollution due to alkylbenzenes. Bell and colleagues (1993)

274

have measured VOCs at various parts of four wastewater treatment plants in Ontario 12 ACS Paragon Plus Environment

Page 13 of 28


275

(Canada) and reported very high emissions of VOCs from aeration basins and

276

process vessels, namely 2700 to 3900 g/d or 36 to 50 g/1000 m3 of wastewater

277

treated, with maximum values of 58 µg/m3 for benzene, 1940 µg/m3 for toluene, 295

278

µg/m3 for ethylbenzene, 953 µg/m3 for (m+p)-xylene, and 460 µg/m3 for o-xylene in

279

the off-gas.24

280

As shown in SI, Table S2, and other studies,11, 25, 26 benzene is highly spatially

281

correlated with other BTEX species. All LUR studies on alkylbenzenes have modeled

282

at least benzene as a marker of toxic air pollutants. However, Aguilera et al. (2008)

283

only modeled TBTEX, which is the sum of all BTEX pollutants.27 Three LUR studies

284

have modeled only benzene, three others modeled all BTEX species, and some

285

others modeled a few VOCs, which could be due to limitations in measurements or

286

the aims and/or interests of the research groups. As we were interested not only in

287

the exposure assessment for epidemiologic research purposes but also in

288

understanding the individual predictors for air quality management and monitoring,

289

we modeled all measured pollutants separately. A few relevant distinctive results

290

were found for Tehran.

291

The benzene model was mainly driven by surrogates of traffic, such as length of all

292

roads in a buffer of 5000 m, log-distance to taxi lines, ancillary roads in a buffer of 50

293

m, taxi lines in a buffer of 25 m, and distance to bus terminals. However, distance to

294

sewage treatment plants, official/commercial land use areas, and sensitive areas also

295

explained part of the variability. Previous studies on taxi drivers in central heavily

296

polluted districts of Tehran showed an increased level of chromosome aberration in

297

comparison with taxi drivers of non-polluted areas (in Lahijan city, Iran).28 This might

298

be associated with the high level of these toxic pollutants in central areas of Tehran 13 ACS Paragon Plus Environment


299

and along the major taxi lines (Figure 3). Bus terminals were also explaining long-

300

term variation of criteria air pollutants in our previous analyses14, 15 and need special

301

attention for air quality management given their presence in benzene and m-xylene

302

models too. In other contexts, most of the benzene variability has also been

303

explained by traffic-related variables, especially length of roads or highways in small

304

buffer sizes (e.g. 50 to 100 m) but these studies usually have not considered large

305

buffer sizes up to 5000 m.10, 29 Our study suggests that large buffers are needed to

306

explain local annual mean concentrations of benzene in complex situations of a

307

megacity like Tehran. The commercial land use areas both in Tehran (in a buffer of

308

2500 m) and in Toronto (in a buffer of 2900 m)10 explained part of the benzene

309

variability. The R2 values for modelled benzene ranged from 0.43 for Detroit (USA) 25

310

to 0.93 for El Paso (USA).30 It has been 0.67 in Toronto (population of 2.7 millions)

311

and 0.65 in New York City (population of 8.4 millions) while we reached an R2 value

312

of 0.70 in Tehran (Table 2).

313

The predictors of toluene were similar to those of benzene in Tehran but there were

314

some other variables that explained toluene variability, such as log-distance to gas

315

filling stores, log-distance to alleys, and ratio of green space areas in buffer of 5000

316

m and distance to green spaces. Few LUR studies on VOCs have modeled toluene,

317

with R2 ranging from 0.31 for Detroit25 to 0.81 for Sarina (Canada).11 Some LUR

318

models have reported low to moderate R2 values for toluene, e.g. 0.41 in Dallas

319

(USA),26 0.46 in Windsor (Canada),29 and 0.54 in Rome (Italy).31 However, some

320

studies reported high R2 values for toluene, e.g. 0.76 in Munich (Germany)32 and

321

0.79 in Ottawa (Canada).9 In Sarnia, in addition to traffic-related variables, toluene

322

was mainly explained by industrial area in a buffer of 2800 m.11 In Ottawa, National

323

Pollutant Release Inventory VOC facility count in a 4 km buffer and National Pollutant 14 ACS Paragon Plus Environment

Page 14 of 28

Page 15 of 28


324

Release Inventory toluene facility count in a 8 km buffer explained variability of

325

toluene along with the traffic-related predictors.9 In Tehran megacity we could explain

326

65% of toluene spatial variability by the available predictors (Table 2).

327

The ethylbenzene model in Tehran, in addition to the predictor variables shared with

328

benzene and toluene, had some other explanatory variables. These were industrial

329

land use areas in a buffer of 5000 m and residential land use areas in a buffer of

330

3500 m. Overall, five LUR studies for VOCs have modeled ethylbenzene.10, 11, 25, 26, 32

331

The R2 values of these 5 studies have been 0.40 to 0.63 in seasonal models of

332

Dallas,26 0.63 in Detroit,25 0.67 in a National Canadian model,33 0.79 in Munich,32 and

333

0.81 in Sarnia.11 In Tehran megacity we could explain 66% of ethylbenzene spatial

334

variability (Table 2). The predictors of ethylbenzene were similar to those of benzene

335

and toluene in other studies.

336

Overall, four LUR studies for VOCs have developed models for xylenes or (m+p)-

337

xylene but none has modeled m-xylene and p-xylene separately.9, 11, 25, 26 The m-

338

xylene and p-xylene models in Tehran showed similarity with benzene, toluene, and

339

ethylbenzene models but o-xylene behaved a little differently in modeling. It showed

340

the lowest correlation with benzene, with a spearman rank correlation coefficient of

341

0.81 (SI, Table S2). In fact, not all roads in a buffer of 5000 m but official/commercial

342

land use area in buffer of 3500 m was the starting model for o-xylene and three other

343

predictors, namely distance to agricultural land use, block density in a buffer of 250

344

m, and distance to educational areas, were selected into the final model in addition to

345

predictors shared with other pollutants. Three other studies that modeled o-xylene, to

346

date, found R2 values of 0.37 to 0.60 (seasonal models in Dallas),26 0.60 (Detroit),25



Page 16 of 28

347

and 0.80 (Sarnia)11 while we could explain 0.66 of long-term spatial variability in

348

Tehran.

349

So far four LUR studies on VOCs have developed models for TBTEX.11, 12, 25, 27 The

350

R2 values in these studies have been 0.40 for Detroit,25 0.70 for New York,12 0.74 for

351

Sabadell,27 and 0.81 for Sarnia,11 and in Tehran we could explain 0.66 of the

352

variation in the observed values. The predictors of TBTEX were a range of variables

353

mostly shared with other BTEX pollutants.

354

In light of the Tehran topography and wind patterns, we anticipated weather factors

355

to be potentially influential determinants of the spatial distribution of BTEX. Indeed,

356

our univariate regression analyses showed that maximum adjusted-R2 for WRF-

357

based predictors ranged from 0.15 for p-xylene to 0.21 for toluene. However, in our

358

multivariate spatial models, WRF-based predictors turned out to be sufficiently

359

captured by other main predictors, thus, none of these weather parameters ended up

360

in the final models. To date, no LUR study on VOCs has incorporated long-term

361

meteorological variables in the modeling; however, depending on the availability of

362

spatial data and local conditions, these variables may be relevant predictors in other

363

cities.

364

Most interestingly, the Tehran VOC emission inventory data did not sufficiently

365

explain spatial variability of any alkylbenzenes to be included in the final models

366

despite significant univariate correlations of several emission inventory data with our

367

measurements. This finding is in contrast to the LUR studies on VOCs in Windsor,29

368

Detroit,25 Ottawa9, a National Canadian model,33 but in line with a NO2 LUR model in

369

Rome (Italy)34. The collinearity between emissions and the set of spatial

370

determinants used in our final models may explain this finding. Not to rely on 16 ACS Paragon Plus Environment

Page 17 of 28


371

emission inventory data for the spatial modelling of VOCs might be an advantage in

372

cities where those data are not available or where quality of inventories is uncertain.

373

The average VIF of variables, as a measure of multicollinearity, was less than 2.2 in

374

all models with a maximum VIF of 5.6 for ethylbenzene model. Though there is no

375

consensus in the LUR community about the VIF, we have used a VIF of smaller than

376

10 as rule of thumb for our modeling as recommended by Kutner et al..14,

377

Moran’s I values in our residuals were close to zero, suggesting that our models

378

captured the relevant spatial patterns of average alkylbenzene levels.

379

Our

380

disadvantaged areas of Tehran showed somewhat higher concentrations for the

381

latter one. However, on average, these differences are rather small. To better

382

understand the difference in home outdoor exposure between wealthy and

383

disadvantaged people, more specific analyses are needed linking people’s residential

384

location and socio-economic status (SES) to our map. This will reveal to what extent

385

SES determines exposure on the local neighborhood scale.

386

Benzene, as a genotoxic carcinogen has no safe level of exposure and for leukemia,

387

the World Health Organization (WHO) guidelines associate 0.17 µg/m3 exposure to

388

airborne benzene with an excess lifetime risk of 1/1,000,000.35 Based on this unit

389

risk, the life time exposure to ambient annual mean benzene concentrations in

390

Tehran translates into 49.4 cases per 1,000,000 or about 445 leukemia cases per 9

391

million residents of Tehran. Based on the unit risk used by the U.S. Environmental

392

Protection Agency (7.8×10-6 per one µg/m3) the estimate for Tehran results in about

393

600 cases.36

analyses

on

differences

between

wealthier

advantaged

and

22

The

poorer



Page 18 of 28

394

We have derived high resolution models of alkylbenzenes in Tehran. These

395

estimates could be used for various health effects studies, urban planning, air quality

396

management, and the monitoring of evidence-based policy making. We have recently

397

systematically reviewed all LUR models for VOCs,37 and found that this study is the

398

largest to predict all BTEX species in a megacity.

399 400

ASSOCIATED CONTENT

401

Supplemental Information

402

The document contains 11 additional pages, 4 additional tables and 4 additional

403

figures including:

404

Table S1. All 520 variables and respective buffer sizes, assumed direction of the

405

effect, and units derived for the use in LUR models.

406

Table S2. Spatial spearman correlations of derived annual means.

407

Table S3. The Spearman correlations of predicted annual means.

408

Table S4. Summary statistics of predicted alkylbenzenes concentrations over

409

advantaged (north to center) and disadvantaged (south to center) areas.

410

Figure S1. Predicted annual mean benzene, toluene, and ethylbenzene.

411

Figure S2. Predicted annual mean p-xylene, m-xylene, o-xylene, and Total BTEX.

412

Figure S3. The agreement of measured versus predicted concentrations.

413

Figure S4. Predicted annual mean benzene concentrations over advantaged versus

414

disadvantaged areas.

415

The SI is available free of charge via the Internet at http://pubs.acs.org.


Page 19 of 28


416

ORCID IDs:

417

Heresh Amini: 0000-0002-4825-1322

418

Vahid Hosseini: 0000-0001-8977-7016

419

Masud Yunesian: 0000-0002-2870-7433

420

Nino Künzli: 0000-0001-8360-080X

421 422

ACKNOWLEDGMENT

423

The authors acknowledge Tehran Air Quality Control Company staff and numerous

424

students who contributed in data collection of Tehran SEPEHR. Heresh Amini

425

sincerely acknowledges the Swiss Government, the Swiss Tropical and Public Health

426

Institute, the Swiss School of Public Health (SSPH+) and the PhD Program Health

427

Sciences (PPHS) for their superb academic support.

428

Funding: Tehran Air Quality Control Company, a subsidiary of Tehran Municipality,

429

funded the Tehran SEPEHR project and so we acknowledge this gesture. Heresh

430

Amini holds a Swiss Government Excellence Scholarship (ESKAS) for PhD in

431

Epidemiology and a project stipend of the PhD Program Health Sciences (PPHS) at

432

the Faculty of Medicine, the University of Basel, Switzerland.

433

Conflict of Interest: Vahid Hosseini declares that he is affiliated to Tehran Air

434

Quality Control Company (AQCC). The views expressed in this manuscript are those

435

of the author and do not necessarily reflect the views or policies of the Tehran AQCC.

436

The rest of authors declare that they have no actual or potential financial competing

437

interests.



Page 20 of 28

TABLES Table 1. Summary statistics of derived annual means of alkylbenzenes (µg/m3) over 179 measurement sites in Tehran SEPEHR, Tehran, Iran

Benzene Toluene Ethylbenzene p-xylene m-xylene o-xylene Total BTEX

Mean 8.4 24.7 5.9 5.8 10.9 6.3 61.8

SD 3.3 10.8 2.2 2.1 4.6 2.5 25.0

Min 2.1 6.1 1.4 1.7 2.6 2.1 16.4

Max 25.8 88.9 17.8 16.8 34.7 17.8 195.0

th

25 6.3 17.7 4.5 4.6 8.1 4.6 46.0

Percentiles 50th 75th 7.8 9.9 23.2 28.6 5.7 6.8 5.6 6.5 10.4 12.5 6.1 7.3 58.6 70.8

ACS Paragon Plus Environment

20

Page 21 of 28


Table 2. Final land use regression (LUR) models for alkylbenzenes in Tehran SEPEHR, Iran. Model predictors \ Response variable Log_Benzenea Intercept 2.0 (1.0) All road (50b) All road (5000) 2.7e-07 (4.4e-08) Ancillary roads (50) 0.001 (2.5e-04) Arid or undeveloped land use areas (700) Blocks (250) Distance to sewage treatment plants –3.2e-05 (5.7e-06) Distance to agricultural land use areas Distance to all bus terminals –3.1e-05 (1.3e-05) Distance to all bus parking areas Distance to educational areas Distance to urban facilities land use areas Green space land use areas (5000) Highways (250) Highways (50) 0.001 (3.4e-04) Industrial land use areas (3500) Industrial land use areas (5000) Log-distance to alleys Log-distance to gas filling stores Log-distance to official/commercial land use areas Log-distance to taxi lines –0.044 (0.015) Official/commercial land use areas (2000) Official/commercial land use areas (2500) 9.9e-08 (2.4e-08) Official/commercial land use areas (3500) Ratio of alleys in buffer of 5000 m/distance to alleys Ratio of green space areas in buffer of 5000 m/distance to green spaces Ratio of highways in buffer of 50 m/distance to highways Ratio of taxi lines in buffer of 3500 m/distance to taxi lines Residential land use areas (3500) Sensitive land use areas (200) Sensitive land use areas (2000) Sensitive land use areas (400) Sensitive land use areas (500) Sensitive land use areas (700) –3.3e-07 (1.1e-07) Taxi lines (25) 0.003 (0.001) Taxi lines (50) Transportation land use areas (5000) Urban facilities land use (200) Urban facilities land use (3500) 0.70 (0.68) Model R2 (adjusted-R2) 0.66 LOOCVc R2 0.17 Root Mean Square Error 1.5 (2.4) Average (maximum) VIFd Maximum p-value among all variables for each 0.02 model 0.07 Max Cook’s D 0.01 (0.38) Moran’s I (p-value) a Results displayed are regression coefficient (standard error); bBuffer radii in was present in all models.

Log_Toluenea 3.3 (0.2) 0.001 (2.5e-04) 2.6e-07 (5.3e-08)

–3.9e-05 (7.7e-06)

Log_Ethylbenzenea 1.4 (0.1) 9.9e-04 (2.8e-04) 2.1e-07 (7.7e-08)

–3.1e-05 (6.4e-06)

Log_p-xylenea 1.5 (0.1)

Log_m-xylenea 2.2 (0.1)

Log_o-xylenea 1.6 (0.07)

Log_Total BTEXa 4.1 (0.19)

1.9e-07 (4.8e-08) 0.001 (2.7e-04) –4.6e-07 (1.6e-07)

3.6e-07 (5.3e-08) 0.001 (3.1e-04)

0.002 (3.0e-04)

3.1e-07 (4.6e-08) 9.9e-04 (2.9e-04)

–2.1e-05 (6.2e-06)

–2.7e-05 (7.1e-06)

0.004 (0.001) –2.8e-05 (7.1e-06) 2.9e-05 (1.3e-05)

–2.6e-05 (7.0e-06)

–4.1e-05 (1.6e-05) –1.9e-05 (5.9e-06)

–4.1e-05 (8.2e-06) 7.5e-05 (2.8e-05)

3.7e-04 (1.3e-04) –1.2e-08 (5.3e-09) 8.1e-04 (3.6e-04)

–1.7e-08 (6.0e-09) 5.0e-05 (2.4e-05) 0.001 (4.1e-04) 3.6e-08 (1.7e-08)

4.6e-08 (1.1e-08) 0.03 (0.015) –0.07 (0.02) –0.04 (0.014) 2.2e-07 (3.9e-08)

–0.04 (0.02) –0.02 (0.009) –0.05 (0.02) 1.3e-07 (3.4e-08)

1.6e-07 (3.3e-08)

–0.04 (0.02) 1.7e-07 (3.9e-08)

1.6e-07 (3.6e-08) 1.5e-07 (1.6e-08) –3.1e-07 (1.2e-07)

–3.6e-08 (1.2e-08)

–2.2e-08 (1.0e-08)

–2.3e-08 (9.7e-09)

–2.7e-08 (1.1e-08)

0.005 (0.002)

–2.6e-08 (1.1e-08) 0.005 (0.001)

2.8e-06 (1.5e-06) 1.7e-08 (7.7e-09) –2.4e-06 (1.3e-06) –6.9e-08 (1.9e-08)

–6.0e-08 (1.8e-08)

–5.5e-08 (1.7e-08)

–8.5e-07 (3.2e-07) –6.2e-07 (2.1e-07) 0.002 (0.001)

0.003 (0.001) 0.001 (2.3e-04)

0.001 (3.2e-04) 1.3e-08 (5.3e-09)

0.001 (2.5e-04) –1.9e-06 (9.9e-07)

0.65 (0.63) 0.6 0.21 1.4 (1.7) 0.02 0.1 – 0.01 (0.47) meters units; cLeave

0.66 (0.64) 0.61 0.19 2.1 (5.6)

0.64 (0.62) 0.59 0.18 1.4 (2.3)

0.04 0.02 0.08 0.09 – 0.02 (0.43) – 0.02 (0.43) one out cross-validation; dVariance inflation


–4.6e-08 (1.3e-08) 0.66 (0.64) 0.61 0.2 1.5 (2.5)

0.66 (0.63) 0.59 0.19 1.5 (1.9)

0.66 (0.64) 0.61 0.19 1.2 (1.5)

0.05 0.07 0.06 0.07 0.09 0.1 – 0.01 (0.47) – 0.01 (0.46) – 0.01 (0.43) factor. Note that variable “Distance to sewage treatment plants”

21


Page 22 of 28

FIGURES Artwork for Abstract


22

Page 23 of 28


Figure 1. The study area of Tehran SEPEHR, Iran, and location of 179 measurement sites, including the five reference stations. Colors of symbols (labeled with the site ID) relate to different land use categories. The map also visualizes population density.


23


Page 24 of 28

Figure 2. Predicted annual mean concentrations of benzene and o-xylene in Tehran SEPEHR, Tehran, Iran. The insets are categorized by quantiles. For ease of interpretation, the first three classes of benzene are shown by blue color indicating areas where annual mean benzene was below 5 µg/m3 (an air quality standard used in some countries). Although both pollutants were mainly driven by trafficrelated variables, industrial areas, mainly in western part of the city, explained variability of o-xylene (top right panel of the o-xylene). See SI, Figure S1 and Figure S2 for map of all pollutants. The spatial resolution of maps is 5×5 m.


24

Page 25 of 28


Figure 3. Cross-city concentration profiles of predicted annual mean alkylbenzenes (µg/m3) along the transect indicated in the upper right panel by black bar from north to south of Tehran over 28.4 km. As shown, toluene was the predominant pollutant among alkylbenzenes. Spatial peaks all across the profile are mainly caused by main roads or highways. The most polluted area is the city center.


25


Page 26 of 28

REFERENCES 1. Dockery, D. W.; Pope, C. A.; Xu, X. P.; Spengler, J. D.; Ware, J. H.; Fay, M. E.; Ferris, B. G.; Speizer, F. E., An association between air-pollution and mortality in 6 United-States cities. New England Journal of Medicine 1993, 329, (24), 1753-1759. 2. Künzli, N.; Kaiser, R.; Medina, S.; Studnicka, M.; Chanel, O.; Filliger, P.; Herry, M.; Horak, F.; Puybonnieux-Texier, V.; Quenel, P.; Schneider, J.; Seethaler, R.; Vergnaud, J. C.; Sommer, H., Public-health impact of outdoor and traffic-related air pollution: a European assessment. The Lancet 2000, 356, (9232), 795-801. 3. Cohen, A. J.; Brauer, M.; Burnett, R.; Anderson, H. R.; Frostad, J.; Estep, K.; Balakrishnan, K.; Brunekreef, B.; Dandona, L.; Dandona, R.; Feigin, V.; Freedman, G.; Hubbell, B.; Jobling, A.; Kan, H.; Knibbs, L.; Liu, Y.; Martin, R.; Morawska, L.; Pope, C. A., III; Shin, H.; Straif, K.; Shaddick, G.; Thomas, M.; van Dingenen, R.; van Donkelaar, A.; Vos, T.; Murray, C. J. L.; Forouzanfar, M. H., Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015. The Lancet 2017, 389, (10082), 1907-1918. 4. Hoek, G.; Beelen, R.; de Hoogh, K.; Vienneau, D.; Gulliver, J.; Fischer, P.; Briggs, D., A review of landuse regression models to assess spatial variation of outdoor air pollution. Atmospheric Environment 2008, 42, (33), 7561-7578. 5. Heydarpour, P.; Amini, H.; Khoshkish, S.; Seidkhani, H.; Sahraian, M. A.; Yunesian, M., Potential impact of air pollution on multiple sclerosis in Tehran, Iran. Neuroepidemiology 2014, 43, (3-4), 233-238. 6. Jerrett, M.; Arain, A.; Kanaroglou, P.; Beckerman, B.; Potoglou, D.; Sahsuvaroglu, T.; Morrison, J.; Giovis, C., A review and evaluation of intraurban air pollution exposure models. Journal of Exposure Science and Environmental Epidemiology 2005, 15, (2), 185-204. 7. FinlaysonPitts, B. J.; Pitts, J. N., Tropospheric air pollution: Ozone, airborne toxics, polycyclic aromatic hydrocarbons, and particles. Science 1997, 276, (5315), 1045-1052. 8. Baldasano, J. M.; Delgado, R.; Calbo, J., Applying receptor models to analyze urban/suburban VOCs air quality in Martorell (Spain). Environmental Science & Technology 1998, 32, (3), 405-412. 9. Oiamo, T. H.; Johnson, M.; Tang, K.; Luginaah, I. N., Assessing traffic and industrial contributions to ambient nitrogen dioxide and volatile organic compounds in a low pollution urban environment. Science of the Total Environment 2015, 529, 149-157. 10. Su, J. G.; Jerrett, M.; Beckerman, B.; Verma, D.; Arain, M. A.; Kanaroglou, P.; Stieb, D.; Finkelstein, M.; Brook, J., A land use regression model for predicting ambient volatile organic compound concentrations in Toronto, Canada. Atmospheric Environment 2010, 44, (29), 3529-3537. 11. Atari, D. O.; Luginaah, I. N., Assessing the distribution of volatile organic compounds using land use regression in Sarnia, "Chemical Valley", Ontario, Canada. Environmental Health 2009, 8. 12. Kheirbek, I.; Johnson, S.; Ross, Z.; Pezeshki, G.; Ito, K.; Eisl, H.; Matte, T., Spatial variability in levels of benzene, formaldehyde, and total benzene, toluene, ethylbenzene and xylenes in New York City: a land-use regression study. Environmental Health 2012, 11.


26

Page 27 of 28


13. West, J. J.; Cohen, A.; Dentener, F.; Brunekreef, B.; Zhu, T.; Armstrong, B.; Bell, M. L.; Brauer, M.; Carmichael, G.; Costa, D. L., What we breathe impacts our health: improving understanding of the link between air pollution and health. Environmental Science & Technology 2016, 50, (10), 4895–4904. 14. Amini, H.; Taghavi-Shahri, S. M.; Henderson, S. B.; Naddafi, K.; Nabizadeh, R.; Yunesian, M., Land use regression models to estimate the annual and seasonal spatial variability of sulfur dioxide and particulate matter in Tehran, Iran. Science of the Total Environment 2014, 488, 343-353. 15. Amini, H.; Taghavi-Shahri, S. M.; Henderson, S. B.; Hosseini, V.; Hassankhany, H.; Naderi, M.; Ahadi, S.; Schindler, C.; Künzli, N.; Yunesian, M., Annual and seasonal spatial models for nitrogen oxides in Tehran, Iran. Scientific Reports 2016, 6, 11. 16. Amini, H.; Hosseini, V.; Schindler, C.; Hassankhany, H.; Yunesian, M.; Henderson, S. B.; Künzli, N., Spatiotemporal description of BTEX volatile organic compounds in a Middle Eastern megacity: Tehran Study of Exposure Prediction for Environmental Health Research (Tehran SEPEHR). Environmental Pollution 2017, 226, 219-229. 17. Hoek, G.; Beelen, R.; Kos, G.; Dijkema, M.; Zee, S. C. v. d.; Fischer, P. H.; Brunekreef, B., Land use regression model for ultrafine particles in Amsterdam. Environmental Science & Technology 2010, 45, (2), 622628. 18. Shahbazi, H.; Taghvaee, S.; Hosseini, V.; Afshin, H., A GIS based emission inventory development for Tehran. Urban Climate 2016, 17, 216-229. 19. Shahbazi, H.; Reyhanian, M.; Hosseini, V.; Afshin, H., The relative contributions of mobile sources to air pollutant emissions in Tehran, Iran: an emission inventory approach. Emission Control Science and Technology 2016, 2, (1), 44-56. 20. Skamarock, W. C.; Klemp, J. B.; Dudhia, J.; Gill, D. O.; Barker, D. M.; Duda, M. G.; Huang, X.-Y.; Wang, W.; Powers, J. G. A description of the advanced research WRF Version 3, NCAR Technical Note, NCAR/TN45+STR; 2008. 21. Grell, G. A.; Dudhia, J.; Stauffer, D. R. A description of the fifth generation Penn State/NCAR Mesoscale Model (MM5). NCAR Technical Note, NCAR TN-398-STR; 1994. 22.

Kutner MH; Nachtsheim C; Neter J, Applied linear regression models. New York: McGraw-Hill. 2004.

23. Gurung, A.; Levy, J. I.; Bell, M. L., Modeling the intraurban variation in nitrogen dioxide in urban areas in Kathmandu Valley, Nepal. Environmental Research 2017, 155, 42-48. 24. Bell, J.; Melcer, H.; Monteith, H.; Osinga, I.; Steel, P., Stripping of volatile organic compounds at full-scale municipal wastewater treatment plants. Water Environment Research 1993, 65, (6), 708-716. 25. Mukerjee, S.; Smith, L. A.; Johnson, M. M.; Neas, L. M.; Stallings, C. A., Spatial analysis and land use regression of VOCs and NO2 from school-based urban air monitoring in Detroit/Dearborn, USA. Science of The Total Environment 2009, 407, (16), 4642-4651. 26. Smith, L. A.; Mukerjee, S.; Chung, K. C.; Afghani, J., Spatial analysis and land use regression of VOCs and NO2 in Dallas, Texas during two seasons. Journal of Environmental Monitoring 2011, 13, (4), 999-1007.


27


Page 28 of 28

27. Aguilera, I.; Sunyer, J.; Fernandez-Patier, R.; Hoek, G.; Aguirre-Alfaro, A.; Meliefste, K.; BomboiMingarro, M. T.; Nieuwenhuijsen, M. J.; Herce-Garraleta, D.; Brunekreef, B., Estimation of outdoor NOx, NO2, and BTEX exposure in a cohort of pregnant women using land use regression modeling. Environmental Science & Technology 2008, 42, (3), 815-821. 28. Taghizadeh, S.; Najmabadi, H.; Kamali, K.; Behjati, F., Evaluation of chromosomal aberrations caused by air pollutants in some taxi drivers from two polluted districts of urban Tehran and its comparison with drivers from rural areas of Lahijan: a pilot study. Journal of Environmental Health Science and Engineering 2014, 12, 144. 29. Wheeler, A. J.; Smith-Doiron, M.; Xu, X.; Gilbert, N. L.; Brook, J. R., Intra-urban variability of air pollution in Windsor, Ontario - Measurement and modeling for human exposure assessment. Environmental Research 2008, 106, (1), 7-16. 30. Smith, L.; Mukerjee, S.; Gonzales, M.; Stallings, C.; Neas, L.; Norris, G.; Özkaynak, H., Use of GIS and ancillary variables to predict volatile organic compound and nitrogen dioxide levels at unmonitored locations. Atmospheric Environment 2006, 40, (20), 3773-3787. 31. Gaeta, A.; Cattani, G.; di Bucchianico, A. D. M.; De Santis, A.; Cesaroni, G.; Badaloni, C.; Ancona, C.; Forastiere, F.; Sozzi, R.; Bolignano, A.; Sacco, F.; Grp, S. S., Development of nitrogen dioxide and volatile organic compounds land use regression models to estimate air pollution exposure near an Italian airport. Atmospheric Environment 2016, 131, 254-262. 32. Carr, D.; von Ehrenstein, O.; Weiland, S.; Wagner, C.; Wellie, O.; Nicolai, T.; von Mutius, E., Modeling Annual Benzene, Toluene, NO2, and Soot Concentrations on the Basis of Road Traffic Characteristics. Environmental Research 2002, 90, (2), 111-118. 33. Hystad, P.; Setton, E.; Cervantes, A.; Poplawski, K.; Deschenes, S.; Brauer, M.; van Donkelaar, A.; Lamsal, L.; Martin, R.; Jerrett, M.; Demers, P., Creating National Air Pollution Models for Population Exposure Assessment in Canada. Environmental Health Perspectives 2011, 119, (8), 1123-1129. 34. Rosenlund, M.; Forastiere, F.; Stafoggia, M.; Porta, D.; Perucci, M.; Ranzi, A.; Nussio, F.; Perucci, C. A., Comparison of regression models with land-use and emissions data to predict the spatial distribution of trafficrelated air pollution in Rome. Journal of Exposure Science and Environmental Epidemiology 2008, 18, (2), 192199. 35. World Health Organization Regional Office for Europe, WHO guidelines for indoor air quality: selected pollutants. WHO: Bonn, 2010. 36. Loh, M. M.; Levy, J. I.; Spengler, J. D.; Houseman, E. A.; Bennett, D. H., Ranking cancer risks of organic hazardous air pollutants in the United States. Environmental Health Perspectives 2007, 115, (8), 1160-1168. 37. Amini, H.; Yunesian, M.; Hosseini, V.; Schindler, C.; Henderson, S. B.; Künzli, N., A systematic review of land use regression models for volatile organic compounds. Submitted.


28

Land Use Regression Models for Alkylbenzenes in a Middle Eastern

Recommend Documents